]>
Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
2817b000 SW |
2 | |
3 | #include <linux/spinlock.h> | |
4 | #include <linux/fs_struct.h> | |
5 | #include <linux/namei.h> | |
5a0e3ad6 | 6 | #include <linux/slab.h> |
2817b000 SW |
7 | #include <linux/sched.h> |
8 | ||
9 | #include "super.h" | |
3d14c5d2 | 10 | #include "mds_client.h" |
2817b000 SW |
11 | |
12 | /* | |
13 | * Directory operations: readdir, lookup, create, link, unlink, | |
14 | * rename, etc. | |
15 | */ | |
16 | ||
17 | /* | |
18 | * Ceph MDS operations are specified in terms of a base ino and | |
19 | * relative path. Thus, the client can specify an operation on a | |
20 | * specific inode (e.g., a getattr due to fstat(2)), or as a path | |
21 | * relative to, say, the root directory. | |
22 | * | |
23 | * Normally, we limit ourselves to strict inode ops (no path component) | |
24 | * or dentry operations (a single path component relative to an ino). The | |
25 | * exception to this is open_root_dentry(), which will open the mount | |
26 | * point by name. | |
27 | */ | |
28 | ||
29 | const struct inode_operations ceph_dir_iops; | |
30 | const struct file_operations ceph_dir_fops; | |
52dfb8ac | 31 | const struct dentry_operations ceph_dentry_ops; |
2817b000 SW |
32 | |
33 | /* | |
34 | * Initialize ceph dentry state. | |
35 | */ | |
36 | int ceph_init_dentry(struct dentry *dentry) | |
37 | { | |
38 | struct ceph_dentry_info *di; | |
39 | ||
40 | if (dentry->d_fsdata) | |
41 | return 0; | |
42 | ||
36e21687 | 43 | di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO); |
2817b000 SW |
44 | if (!di) |
45 | return -ENOMEM; /* oh well */ | |
46 | ||
47 | spin_lock(&dentry->d_lock); | |
8c6efb58 SW |
48 | if (dentry->d_fsdata) { |
49 | /* lost a race */ | |
50 | kmem_cache_free(ceph_dentry_cachep, di); | |
2817b000 | 51 | goto out_unlock; |
8c6efb58 | 52 | } |
48d0cbd1 | 53 | |
8842b3be | 54 | if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP) |
48d0cbd1 SW |
55 | d_set_d_op(dentry, &ceph_dentry_ops); |
56 | else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR) | |
57 | d_set_d_op(dentry, &ceph_snapdir_dentry_ops); | |
58 | else | |
59 | d_set_d_op(dentry, &ceph_snap_dentry_ops); | |
60 | ||
2817b000 SW |
61 | di->dentry = dentry; |
62 | di->lease_session = NULL; | |
2817b000 | 63 | dentry->d_time = jiffies; |
48d0cbd1 SW |
64 | /* avoid reordering d_fsdata setup so that the check above is safe */ |
65 | smp_mb(); | |
66 | dentry->d_fsdata = di; | |
2817b000 SW |
67 | ceph_dentry_lru_add(dentry); |
68 | out_unlock: | |
69 | spin_unlock(&dentry->d_lock); | |
70 | return 0; | |
71 | } | |
72 | ||
5f21c96d SW |
73 | struct inode *ceph_get_dentry_parent_inode(struct dentry *dentry) |
74 | { | |
75 | struct inode *inode = NULL; | |
76 | ||
77 | if (!dentry) | |
78 | return NULL; | |
79 | ||
80 | spin_lock(&dentry->d_lock); | |
8842b3be | 81 | if (!IS_ROOT(dentry)) { |
5f21c96d SW |
82 | inode = dentry->d_parent->d_inode; |
83 | ihold(inode); | |
84 | } | |
85 | spin_unlock(&dentry->d_lock); | |
86 | return inode; | |
87 | } | |
2817b000 SW |
88 | |
89 | ||
90 | /* | |
91 | * for readdir, we encode the directory frag and offset within that | |
92 | * frag into f_pos. | |
93 | */ | |
94 | static unsigned fpos_frag(loff_t p) | |
95 | { | |
96 | return p >> 32; | |
97 | } | |
98 | static unsigned fpos_off(loff_t p) | |
99 | { | |
100 | return p & 0xffffffff; | |
101 | } | |
102 | ||
4d5f5df6 YZ |
103 | static int fpos_cmp(loff_t l, loff_t r) |
104 | { | |
105 | int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r)); | |
106 | if (v) | |
107 | return v; | |
108 | return (int)(fpos_off(l) - fpos_off(r)); | |
109 | } | |
110 | ||
2817b000 SW |
111 | /* |
112 | * When possible, we try to satisfy a readdir by peeking at the | |
113 | * dcache. We make this work by carefully ordering dentries on | |
114 | * d_u.d_child when we initially get results back from the MDS, and | |
115 | * falling back to a "normal" sync readdir if any dentries in the dir | |
116 | * are dropped. | |
117 | * | |
2f276c51 | 118 | * Complete dir indicates that we have all dentries in the dir. It is |
2817b000 SW |
119 | * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by |
120 | * the MDS if/when the directory is modified). | |
121 | */ | |
a30be7cb YZ |
122 | static int __dcache_readdir(struct file *file, struct dir_context *ctx, |
123 | u32 shared_gen) | |
2817b000 | 124 | { |
77acfa29 AV |
125 | struct ceph_file_info *fi = file->private_data; |
126 | struct dentry *parent = file->f_dentry; | |
2817b000 SW |
127 | struct inode *dir = parent->d_inode; |
128 | struct list_head *p; | |
129 | struct dentry *dentry, *last; | |
130 | struct ceph_dentry_info *di; | |
131 | int err = 0; | |
132 | ||
133 | /* claim ref on last dentry we returned */ | |
134 | last = fi->dentry; | |
135 | fi->dentry = NULL; | |
136 | ||
a30be7cb YZ |
137 | dout("__dcache_readdir %p v%u at %llu (last %p)\n", |
138 | dir, shared_gen, ctx->pos, last); | |
2817b000 | 139 | |
2fd6b7f5 | 140 | spin_lock(&parent->d_lock); |
2817b000 SW |
141 | |
142 | /* start at beginning? */ | |
77acfa29 AV |
143 | if (ctx->pos == 2 || last == NULL || |
144 | ctx->pos < ceph_dentry(last)->offset) { | |
2817b000 SW |
145 | if (list_empty(&parent->d_subdirs)) |
146 | goto out_unlock; | |
147 | p = parent->d_subdirs.prev; | |
148 | dout(" initial p %p/%p\n", p->prev, p->next); | |
149 | } else { | |
150 | p = last->d_u.d_child.prev; | |
151 | } | |
152 | ||
153 | more: | |
154 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
155 | di = ceph_dentry(dentry); | |
156 | while (1) { | |
1cd3935b SW |
157 | dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next, |
158 | d_unhashed(dentry) ? "!hashed" : "hashed", | |
2817b000 SW |
159 | parent->d_subdirs.prev, parent->d_subdirs.next); |
160 | if (p == &parent->d_subdirs) { | |
9cfa1098 | 161 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
162 | goto out_unlock; |
163 | } | |
2fd6b7f5 | 164 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
a30be7cb YZ |
165 | if (di->lease_shared_gen == shared_gen && |
166 | !d_unhashed(dentry) && dentry->d_inode && | |
09b8a7d2 | 167 | ceph_snap(dentry->d_inode) != CEPH_SNAPDIR && |
1d1de916 | 168 | ceph_ino(dentry->d_inode) != CEPH_INO_CEPH && |
4d5f5df6 | 169 | fpos_cmp(ctx->pos, di->offset) <= 0) |
2817b000 SW |
170 | break; |
171 | dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry, | |
172 | dentry->d_name.len, dentry->d_name.name, di->offset, | |
77acfa29 | 173 | ctx->pos, d_unhashed(dentry) ? " unhashed" : "", |
2817b000 | 174 | !dentry->d_inode ? " null" : ""); |
da502956 | 175 | spin_unlock(&dentry->d_lock); |
2817b000 SW |
176 | p = p->prev; |
177 | dentry = list_entry(p, struct dentry, d_u.d_child); | |
178 | di = ceph_dentry(dentry); | |
179 | } | |
180 | ||
da502956 | 181 | dget_dlock(dentry); |
b7ab39f6 | 182 | spin_unlock(&dentry->d_lock); |
2fd6b7f5 | 183 | spin_unlock(&parent->d_lock); |
2817b000 | 184 | |
77acfa29 | 185 | dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, ctx->pos, |
2817b000 | 186 | dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode); |
77acfa29 AV |
187 | ctx->pos = di->offset; |
188 | if (!dir_emit(ctx, dentry->d_name.name, | |
189 | dentry->d_name.len, | |
ad1fee96 | 190 | ceph_translate_ino(dentry->d_sb, dentry->d_inode->i_ino), |
77acfa29 AV |
191 | dentry->d_inode->i_mode >> 12)) { |
192 | if (last) { | |
2817b000 SW |
193 | /* remember our position */ |
194 | fi->dentry = last; | |
f0494206 | 195 | fi->next_offset = fpos_off(di->offset); |
2817b000 | 196 | } |
77acfa29 AV |
197 | dput(dentry); |
198 | return 0; | |
2817b000 | 199 | } |
f5b06628 | 200 | |
77acfa29 AV |
201 | if (last) |
202 | dput(last); | |
203 | last = dentry; | |
2817b000 | 204 | |
77acfa29 | 205 | ctx->pos++; |
2817b000 | 206 | |
b5c84bf6 | 207 | /* make sure a dentry wasn't dropped while we didn't have parent lock */ |
2f276c51 YZ |
208 | if (!ceph_dir_is_complete(dir)) { |
209 | dout(" lost dir complete on %p; falling back to mds\n", dir); | |
efa4c120 SW |
210 | err = -EAGAIN; |
211 | goto out; | |
212 | } | |
213 | ||
2fd6b7f5 | 214 | spin_lock(&parent->d_lock); |
efa4c120 SW |
215 | p = p->prev; /* advance to next dentry */ |
216 | goto more; | |
2817b000 SW |
217 | |
218 | out_unlock: | |
2fd6b7f5 | 219 | spin_unlock(&parent->d_lock); |
efa4c120 SW |
220 | out: |
221 | if (last) | |
2817b000 | 222 | dput(last); |
2817b000 SW |
223 | return err; |
224 | } | |
225 | ||
226 | /* | |
227 | * make note of the last dentry we read, so we can | |
228 | * continue at the same lexicographical point, | |
229 | * regardless of what dir changes take place on the | |
230 | * server. | |
231 | */ | |
232 | static int note_last_dentry(struct ceph_file_info *fi, const char *name, | |
233 | int len) | |
234 | { | |
235 | kfree(fi->last_name); | |
236 | fi->last_name = kmalloc(len+1, GFP_NOFS); | |
237 | if (!fi->last_name) | |
238 | return -ENOMEM; | |
239 | memcpy(fi->last_name, name, len); | |
240 | fi->last_name[len] = 0; | |
241 | dout("note_last_dentry '%s'\n", fi->last_name); | |
242 | return 0; | |
243 | } | |
244 | ||
77acfa29 | 245 | static int ceph_readdir(struct file *file, struct dir_context *ctx) |
2817b000 | 246 | { |
77acfa29 AV |
247 | struct ceph_file_info *fi = file->private_data; |
248 | struct inode *inode = file_inode(file); | |
2817b000 | 249 | struct ceph_inode_info *ci = ceph_inode(inode); |
3d14c5d2 YS |
250 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
251 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
77acfa29 AV |
252 | unsigned frag = fpos_frag(ctx->pos); |
253 | int off = fpos_off(ctx->pos); | |
2817b000 SW |
254 | int err; |
255 | u32 ftype; | |
256 | struct ceph_mds_reply_info_parsed *rinfo; | |
2817b000 | 257 | |
77acfa29 | 258 | dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off); |
9cfa1098 | 259 | if (fi->flags & CEPH_F_ATEND) |
2817b000 SW |
260 | return 0; |
261 | ||
262 | /* always start with . and .. */ | |
77acfa29 | 263 | if (ctx->pos == 0) { |
2817b000 SW |
264 | /* note dir version at start of readdir so we can tell |
265 | * if any dentries get dropped */ | |
2f276c51 | 266 | fi->dir_release_count = atomic_read(&ci->i_release_count); |
2817b000 SW |
267 | |
268 | dout("readdir off 0 -> '.'\n"); | |
77acfa29 | 269 | if (!dir_emit(ctx, ".", 1, |
ad1fee96 | 270 | ceph_translate_ino(inode->i_sb, inode->i_ino), |
77acfa29 | 271 | inode->i_mode >> 12)) |
2817b000 | 272 | return 0; |
77acfa29 | 273 | ctx->pos = 1; |
2817b000 SW |
274 | off = 1; |
275 | } | |
77acfa29 AV |
276 | if (ctx->pos == 1) { |
277 | ino_t ino = parent_ino(file->f_dentry); | |
2817b000 | 278 | dout("readdir off 1 -> '..'\n"); |
77acfa29 | 279 | if (!dir_emit(ctx, "..", 2, |
ad1fee96 | 280 | ceph_translate_ino(inode->i_sb, ino), |
77acfa29 | 281 | inode->i_mode >> 12)) |
2817b000 | 282 | return 0; |
77acfa29 | 283 | ctx->pos = 2; |
2817b000 SW |
284 | off = 2; |
285 | } | |
286 | ||
287 | /* can we use the dcache? */ | |
be655596 | 288 | spin_lock(&ci->i_ceph_lock); |
77acfa29 | 289 | if ((ctx->pos == 2 || fi->dentry) && |
3d14c5d2 | 290 | !ceph_test_mount_opt(fsc, NOASYNCREADDIR) && |
a0dff78d | 291 | ceph_snap(inode) != CEPH_SNAPDIR && |
2f276c51 | 292 | __ceph_dir_is_complete(ci) && |
2817b000 | 293 | __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) { |
a30be7cb | 294 | u32 shared_gen = ci->i_shared_gen; |
be655596 | 295 | spin_unlock(&ci->i_ceph_lock); |
a30be7cb | 296 | err = __dcache_readdir(file, ctx, shared_gen); |
efa4c120 | 297 | if (err != -EAGAIN) |
2817b000 | 298 | return err; |
efa4c120 | 299 | } else { |
be655596 | 300 | spin_unlock(&ci->i_ceph_lock); |
2817b000 | 301 | } |
2817b000 SW |
302 | if (fi->dentry) { |
303 | err = note_last_dentry(fi, fi->dentry->d_name.name, | |
304 | fi->dentry->d_name.len); | |
305 | if (err) | |
306 | return err; | |
307 | dput(fi->dentry); | |
308 | fi->dentry = NULL; | |
309 | } | |
310 | ||
311 | /* proceed with a normal readdir */ | |
312 | ||
313 | more: | |
314 | /* do we have the correct frag content buffered? */ | |
315 | if (fi->frag != frag || fi->last_readdir == NULL) { | |
316 | struct ceph_mds_request *req; | |
317 | int op = ceph_snap(inode) == CEPH_SNAPDIR ? | |
318 | CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR; | |
319 | ||
320 | /* discard old result, if any */ | |
393f6620 | 321 | if (fi->last_readdir) { |
2817b000 | 322 | ceph_mdsc_put_request(fi->last_readdir); |
393f6620 SW |
323 | fi->last_readdir = NULL; |
324 | } | |
2817b000 | 325 | |
2817b000 SW |
326 | dout("readdir fetching %llx.%llx frag %x offset '%s'\n", |
327 | ceph_vinop(inode), frag, fi->last_name); | |
328 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
329 | if (IS_ERR(req)) | |
330 | return PTR_ERR(req); | |
54008399 YZ |
331 | err = ceph_alloc_readdir_reply_buffer(req, inode); |
332 | if (err) { | |
333 | ceph_mdsc_put_request(req); | |
334 | return err; | |
335 | } | |
70b666c3 SW |
336 | req->r_inode = inode; |
337 | ihold(inode); | |
77acfa29 | 338 | req->r_dentry = dget(file->f_dentry); |
2817b000 SW |
339 | /* hints to request -> mds selection code */ |
340 | req->r_direct_mode = USE_AUTH_MDS; | |
341 | req->r_direct_hash = ceph_frag_value(frag); | |
342 | req->r_direct_is_hash = true; | |
343 | req->r_path2 = kstrdup(fi->last_name, GFP_NOFS); | |
344 | req->r_readdir_offset = fi->next_offset; | |
345 | req->r_args.readdir.frag = cpu_to_le32(frag); | |
2817b000 SW |
346 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
347 | if (err < 0) { | |
348 | ceph_mdsc_put_request(req); | |
349 | return err; | |
350 | } | |
351 | dout("readdir got and parsed readdir result=%d" | |
352 | " on frag %x, end=%d, complete=%d\n", err, frag, | |
353 | (int)req->r_reply_info.dir_end, | |
354 | (int)req->r_reply_info.dir_complete); | |
355 | ||
356 | if (!req->r_did_prepopulate) { | |
357 | dout("readdir !did_prepopulate"); | |
2f276c51 YZ |
358 | /* preclude from marking dir complete */ |
359 | fi->dir_release_count--; | |
2817b000 SW |
360 | } |
361 | ||
362 | /* note next offset and last dentry name */ | |
81c6aea5 YZ |
363 | rinfo = &req->r_reply_info; |
364 | if (le32_to_cpu(rinfo->dir_dir->frag) != frag) { | |
365 | frag = le32_to_cpu(rinfo->dir_dir->frag); | |
366 | if (ceph_frag_is_leftmost(frag)) | |
367 | fi->next_offset = 2; | |
368 | else | |
369 | fi->next_offset = 0; | |
370 | off = fi->next_offset; | |
371 | } | |
f0494206 | 372 | fi->frag = frag; |
2817b000 SW |
373 | fi->offset = fi->next_offset; |
374 | fi->last_readdir = req; | |
375 | ||
376 | if (req->r_reply_info.dir_end) { | |
377 | kfree(fi->last_name); | |
378 | fi->last_name = NULL; | |
7b88dadc SW |
379 | if (ceph_frag_is_rightmost(frag)) |
380 | fi->next_offset = 2; | |
381 | else | |
382 | fi->next_offset = 0; | |
2817b000 | 383 | } else { |
2817b000 SW |
384 | err = note_last_dentry(fi, |
385 | rinfo->dir_dname[rinfo->dir_nr-1], | |
386 | rinfo->dir_dname_len[rinfo->dir_nr-1]); | |
387 | if (err) | |
388 | return err; | |
389 | fi->next_offset += rinfo->dir_nr; | |
390 | } | |
391 | } | |
392 | ||
393 | rinfo = &fi->last_readdir->r_reply_info; | |
394 | dout("readdir frag %x num %d off %d chunkoff %d\n", frag, | |
395 | rinfo->dir_nr, off, fi->offset); | |
77acfa29 AV |
396 | |
397 | ctx->pos = ceph_make_fpos(frag, off); | |
da39822c | 398 | while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) { |
2817b000 SW |
399 | struct ceph_mds_reply_inode *in = |
400 | rinfo->dir_in[off - fi->offset].in; | |
3105c19c SW |
401 | struct ceph_vino vino; |
402 | ino_t ino; | |
403 | ||
2817b000 | 404 | dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n", |
77acfa29 | 405 | off, off - fi->offset, rinfo->dir_nr, ctx->pos, |
2817b000 SW |
406 | rinfo->dir_dname_len[off - fi->offset], |
407 | rinfo->dir_dname[off - fi->offset], in); | |
408 | BUG_ON(!in); | |
409 | ftype = le32_to_cpu(in->mode) >> 12; | |
3105c19c SW |
410 | vino.ino = le64_to_cpu(in->ino); |
411 | vino.snap = le64_to_cpu(in->snapid); | |
412 | ino = ceph_vino_to_ino(vino); | |
77acfa29 | 413 | if (!dir_emit(ctx, |
2817b000 SW |
414 | rinfo->dir_dname[off - fi->offset], |
415 | rinfo->dir_dname_len[off - fi->offset], | |
77acfa29 | 416 | ceph_translate_ino(inode->i_sb, ino), ftype)) { |
2817b000 SW |
417 | dout("filldir stopping us...\n"); |
418 | return 0; | |
419 | } | |
420 | off++; | |
77acfa29 | 421 | ctx->pos++; |
2817b000 SW |
422 | } |
423 | ||
424 | if (fi->last_name) { | |
425 | ceph_mdsc_put_request(fi->last_readdir); | |
426 | fi->last_readdir = NULL; | |
427 | goto more; | |
428 | } | |
429 | ||
430 | /* more frags? */ | |
431 | if (!ceph_frag_is_rightmost(frag)) { | |
432 | frag = ceph_frag_next(frag); | |
433 | off = 0; | |
77acfa29 | 434 | ctx->pos = ceph_make_fpos(frag, off); |
2817b000 SW |
435 | dout("readdir next frag is %x\n", frag); |
436 | goto more; | |
437 | } | |
9cfa1098 | 438 | fi->flags |= CEPH_F_ATEND; |
2817b000 SW |
439 | |
440 | /* | |
441 | * if dir_release_count still matches the dir, no dentries | |
442 | * were released during the whole readdir, and we should have | |
443 | * the complete dir contents in our cache. | |
444 | */ | |
be655596 | 445 | spin_lock(&ci->i_ceph_lock); |
2f276c51 | 446 | if (atomic_read(&ci->i_release_count) == fi->dir_release_count) { |
a8673d61 | 447 | dout(" marking %p complete\n", inode); |
2f276c51 | 448 | __ceph_dir_set_complete(ci, fi->dir_release_count); |
77acfa29 | 449 | ci->i_max_offset = ctx->pos; |
2817b000 | 450 | } |
be655596 | 451 | spin_unlock(&ci->i_ceph_lock); |
2817b000 | 452 | |
77acfa29 | 453 | dout("readdir %p file %p done.\n", inode, file); |
2817b000 SW |
454 | return 0; |
455 | } | |
456 | ||
dcd3cc05 | 457 | static void reset_readdir(struct ceph_file_info *fi, unsigned frag) |
2817b000 SW |
458 | { |
459 | if (fi->last_readdir) { | |
460 | ceph_mdsc_put_request(fi->last_readdir); | |
461 | fi->last_readdir = NULL; | |
462 | } | |
463 | kfree(fi->last_name); | |
a1629c3b | 464 | fi->last_name = NULL; |
dcd3cc05 YZ |
465 | if (ceph_frag_is_leftmost(frag)) |
466 | fi->next_offset = 2; /* compensate for . and .. */ | |
467 | else | |
468 | fi->next_offset = 0; | |
2817b000 SW |
469 | if (fi->dentry) { |
470 | dput(fi->dentry); | |
471 | fi->dentry = NULL; | |
472 | } | |
9cfa1098 | 473 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
474 | } |
475 | ||
965c8e59 | 476 | static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence) |
2817b000 SW |
477 | { |
478 | struct ceph_file_info *fi = file->private_data; | |
479 | struct inode *inode = file->f_mapping->host; | |
f0494206 | 480 | loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset); |
2817b000 SW |
481 | loff_t retval; |
482 | ||
483 | mutex_lock(&inode->i_mutex); | |
06222e49 | 484 | retval = -EINVAL; |
965c8e59 | 485 | switch (whence) { |
2817b000 SW |
486 | case SEEK_END: |
487 | offset += inode->i_size + 2; /* FIXME */ | |
488 | break; | |
489 | case SEEK_CUR: | |
490 | offset += file->f_pos; | |
06222e49 JB |
491 | case SEEK_SET: |
492 | break; | |
493 | default: | |
494 | goto out; | |
2817b000 | 495 | } |
06222e49 | 496 | |
f0494206 | 497 | if (offset >= 0) { |
2817b000 SW |
498 | if (offset != file->f_pos) { |
499 | file->f_pos = offset; | |
500 | file->f_version = 0; | |
9cfa1098 | 501 | fi->flags &= ~CEPH_F_ATEND; |
2817b000 SW |
502 | } |
503 | retval = offset; | |
504 | ||
505 | /* | |
506 | * discard buffered readdir content on seekdir(0), or | |
507 | * seek to new frag, or seek prior to current chunk. | |
508 | */ | |
509 | if (offset == 0 || | |
f0494206 | 510 | fpos_frag(offset) != fi->frag || |
2817b000 SW |
511 | fpos_off(offset) < fi->offset) { |
512 | dout("dir_llseek dropping %p content\n", file); | |
dcd3cc05 | 513 | reset_readdir(fi, fpos_frag(offset)); |
2817b000 SW |
514 | } |
515 | ||
516 | /* bump dir_release_count if we did a forward seek */ | |
f0494206 | 517 | if (fpos_cmp(offset, old_offset) > 0) |
2817b000 SW |
518 | fi->dir_release_count--; |
519 | } | |
06222e49 | 520 | out: |
2817b000 SW |
521 | mutex_unlock(&inode->i_mutex); |
522 | return retval; | |
523 | } | |
524 | ||
525 | /* | |
468640e3 | 526 | * Handle lookups for the hidden .snap directory. |
2817b000 | 527 | */ |
468640e3 SW |
528 | int ceph_handle_snapdir(struct ceph_mds_request *req, |
529 | struct dentry *dentry, int err) | |
2817b000 | 530 | { |
3d14c5d2 | 531 | struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb); |
d79698da | 532 | struct inode *parent = dentry->d_parent->d_inode; /* we hold i_mutex */ |
2817b000 SW |
533 | |
534 | /* .snap dir? */ | |
535 | if (err == -ENOENT && | |
455cec0a | 536 | ceph_snap(parent) == CEPH_NOSNAP && |
6b805185 | 537 | strcmp(dentry->d_name.name, |
3d14c5d2 | 538 | fsc->mount_options->snapdir_name) == 0) { |
2817b000 SW |
539 | struct inode *inode = ceph_get_snapdir(parent); |
540 | dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n", | |
541 | dentry, dentry->d_name.len, dentry->d_name.name, inode); | |
9358c6d4 | 542 | BUG_ON(!d_unhashed(dentry)); |
2817b000 SW |
543 | d_add(dentry, inode); |
544 | err = 0; | |
545 | } | |
468640e3 SW |
546 | return err; |
547 | } | |
2817b000 | 548 | |
468640e3 SW |
549 | /* |
550 | * Figure out final result of a lookup/open request. | |
551 | * | |
552 | * Mainly, make sure we return the final req->r_dentry (if it already | |
553 | * existed) in place of the original VFS-provided dentry when they | |
554 | * differ. | |
555 | * | |
556 | * Gracefully handle the case where the MDS replies with -ENOENT and | |
557 | * no trace (which it may do, at its discretion, e.g., if it doesn't | |
558 | * care to issue a lease on the negative dentry). | |
559 | */ | |
560 | struct dentry *ceph_finish_lookup(struct ceph_mds_request *req, | |
561 | struct dentry *dentry, int err) | |
562 | { | |
2817b000 SW |
563 | if (err == -ENOENT) { |
564 | /* no trace? */ | |
565 | err = 0; | |
566 | if (!req->r_reply_info.head->is_dentry) { | |
567 | dout("ENOENT and no trace, dentry %p inode %p\n", | |
568 | dentry, dentry->d_inode); | |
569 | if (dentry->d_inode) { | |
570 | d_drop(dentry); | |
571 | err = -ENOENT; | |
572 | } else { | |
573 | d_add(dentry, NULL); | |
574 | } | |
575 | } | |
576 | } | |
577 | if (err) | |
578 | dentry = ERR_PTR(err); | |
579 | else if (dentry != req->r_dentry) | |
580 | dentry = dget(req->r_dentry); /* we got spliced */ | |
581 | else | |
582 | dentry = NULL; | |
583 | return dentry; | |
584 | } | |
585 | ||
1d1de916 SW |
586 | static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry) |
587 | { | |
588 | return ceph_ino(inode) == CEPH_INO_ROOT && | |
589 | strncmp(dentry->d_name.name, ".ceph", 5) == 0; | |
590 | } | |
591 | ||
2817b000 SW |
592 | /* |
593 | * Look up a single dir entry. If there is a lookup intent, inform | |
594 | * the MDS so that it gets our 'caps wanted' value in a single op. | |
595 | */ | |
596 | static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry, | |
00cd8dd3 | 597 | unsigned int flags) |
2817b000 | 598 | { |
3d14c5d2 YS |
599 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
600 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
601 | struct ceph_mds_request *req; |
602 | int op; | |
603 | int err; | |
604 | ||
605 | dout("lookup %p dentry %p '%.*s'\n", | |
606 | dir, dentry, dentry->d_name.len, dentry->d_name.name); | |
607 | ||
608 | if (dentry->d_name.len > NAME_MAX) | |
609 | return ERR_PTR(-ENAMETOOLONG); | |
610 | ||
611 | err = ceph_init_dentry(dentry); | |
612 | if (err < 0) | |
613 | return ERR_PTR(err); | |
614 | ||
2817b000 SW |
615 | /* can we conclude ENOENT locally? */ |
616 | if (dentry->d_inode == NULL) { | |
617 | struct ceph_inode_info *ci = ceph_inode(dir); | |
618 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
619 | ||
be655596 | 620 | spin_lock(&ci->i_ceph_lock); |
2817b000 SW |
621 | dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags); |
622 | if (strncmp(dentry->d_name.name, | |
3d14c5d2 | 623 | fsc->mount_options->snapdir_name, |
2817b000 | 624 | dentry->d_name.len) && |
1d1de916 | 625 | !is_root_ceph_dentry(dir, dentry) && |
2f276c51 | 626 | __ceph_dir_is_complete(ci) && |
2817b000 | 627 | (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) { |
be655596 | 628 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
629 | dout(" dir %p complete, -ENOENT\n", dir); |
630 | d_add(dentry, NULL); | |
631 | di->lease_shared_gen = ci->i_shared_gen; | |
632 | return NULL; | |
633 | } | |
be655596 | 634 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
635 | } |
636 | ||
637 | op = ceph_snap(dir) == CEPH_SNAPDIR ? | |
638 | CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
639 | req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS); | |
640 | if (IS_ERR(req)) | |
7e34bc52 | 641 | return ERR_CAST(req); |
2817b000 SW |
642 | req->r_dentry = dget(dentry); |
643 | req->r_num_caps = 2; | |
644 | /* we only need inode linkage */ | |
645 | req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE); | |
646 | req->r_locked_dir = dir; | |
647 | err = ceph_mdsc_do_request(mdsc, NULL, req); | |
468640e3 | 648 | err = ceph_handle_snapdir(req, dentry, err); |
2817b000 SW |
649 | dentry = ceph_finish_lookup(req, dentry, err); |
650 | ceph_mdsc_put_request(req); /* will dput(dentry) */ | |
651 | dout("lookup result=%p\n", dentry); | |
652 | return dentry; | |
653 | } | |
654 | ||
655 | /* | |
656 | * If we do a create but get no trace back from the MDS, follow up with | |
657 | * a lookup (the VFS expects us to link up the provided dentry). | |
658 | */ | |
659 | int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry) | |
660 | { | |
00cd8dd3 | 661 | struct dentry *result = ceph_lookup(dir, dentry, 0); |
2817b000 SW |
662 | |
663 | if (result && !IS_ERR(result)) { | |
664 | /* | |
665 | * We created the item, then did a lookup, and found | |
666 | * it was already linked to another inode we already | |
667 | * had in our cache (and thus got spliced). Link our | |
668 | * dentry to that inode, but don't hash it, just in | |
669 | * case the VFS wants to dereference it. | |
670 | */ | |
671 | BUG_ON(!result->d_inode); | |
672 | d_instantiate(dentry, result->d_inode); | |
673 | return 0; | |
674 | } | |
675 | return PTR_ERR(result); | |
676 | } | |
677 | ||
678 | static int ceph_mknod(struct inode *dir, struct dentry *dentry, | |
1a67aafb | 679 | umode_t mode, dev_t rdev) |
2817b000 | 680 | { |
3d14c5d2 YS |
681 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
682 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
683 | struct ceph_mds_request *req; |
684 | int err; | |
685 | ||
686 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
687 | return -EROFS; | |
688 | ||
1a67aafb | 689 | dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n", |
2817b000 SW |
690 | dir, dentry, mode, rdev); |
691 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS); | |
692 | if (IS_ERR(req)) { | |
693 | d_drop(dentry); | |
694 | return PTR_ERR(req); | |
695 | } | |
696 | req->r_dentry = dget(dentry); | |
697 | req->r_num_caps = 2; | |
698 | req->r_locked_dir = dir; | |
699 | req->r_args.mknod.mode = cpu_to_le32(mode); | |
700 | req->r_args.mknod.rdev = cpu_to_le32(rdev); | |
701 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
702 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
703 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
704 | if (!err && !req->r_reply_info.head->is_dentry) | |
705 | err = ceph_handle_notrace_create(dir, dentry); | |
706 | ceph_mdsc_put_request(req); | |
7221fe4c GZ |
707 | |
708 | if (!err) | |
b20a95a0 YZ |
709 | ceph_init_acl(dentry, dentry->d_inode, dir); |
710 | else | |
2817b000 SW |
711 | d_drop(dentry); |
712 | return err; | |
713 | } | |
714 | ||
4acdaf27 | 715 | static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode, |
ebfc3b49 | 716 | bool excl) |
2817b000 | 717 | { |
2d83bde9 | 718 | return ceph_mknod(dir, dentry, mode, 0); |
2817b000 SW |
719 | } |
720 | ||
721 | static int ceph_symlink(struct inode *dir, struct dentry *dentry, | |
722 | const char *dest) | |
723 | { | |
3d14c5d2 YS |
724 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
725 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
726 | struct ceph_mds_request *req; |
727 | int err; | |
728 | ||
729 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
730 | return -EROFS; | |
731 | ||
732 | dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest); | |
733 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS); | |
734 | if (IS_ERR(req)) { | |
735 | d_drop(dentry); | |
736 | return PTR_ERR(req); | |
737 | } | |
738 | req->r_dentry = dget(dentry); | |
739 | req->r_num_caps = 2; | |
740 | req->r_path2 = kstrdup(dest, GFP_NOFS); | |
741 | req->r_locked_dir = dir; | |
742 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
743 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
744 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
745 | if (!err && !req->r_reply_info.head->is_dentry) | |
746 | err = ceph_handle_notrace_create(dir, dentry); | |
747 | ceph_mdsc_put_request(req); | |
b20a95a0 YZ |
748 | if (!err) |
749 | ceph_init_acl(dentry, dentry->d_inode, dir); | |
750 | else | |
2817b000 SW |
751 | d_drop(dentry); |
752 | return err; | |
753 | } | |
754 | ||
18bb1db3 | 755 | static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
2817b000 | 756 | { |
3d14c5d2 YS |
757 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
758 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
759 | struct ceph_mds_request *req; |
760 | int err = -EROFS; | |
761 | int op; | |
762 | ||
763 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
764 | /* mkdir .snap/foo is a MKSNAP */ | |
765 | op = CEPH_MDS_OP_MKSNAP; | |
766 | dout("mksnap dir %p snap '%.*s' dn %p\n", dir, | |
767 | dentry->d_name.len, dentry->d_name.name, dentry); | |
768 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
18bb1db3 | 769 | dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode); |
2817b000 SW |
770 | op = CEPH_MDS_OP_MKDIR; |
771 | } else { | |
772 | goto out; | |
773 | } | |
774 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
775 | if (IS_ERR(req)) { | |
776 | err = PTR_ERR(req); | |
777 | goto out; | |
778 | } | |
779 | ||
780 | req->r_dentry = dget(dentry); | |
781 | req->r_num_caps = 2; | |
782 | req->r_locked_dir = dir; | |
783 | req->r_args.mkdir.mode = cpu_to_le32(mode); | |
784 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
785 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
786 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
787 | if (!err && !req->r_reply_info.head->is_dentry) | |
788 | err = ceph_handle_notrace_create(dir, dentry); | |
789 | ceph_mdsc_put_request(req); | |
790 | out: | |
b20a95a0 YZ |
791 | if (!err) |
792 | ceph_init_acl(dentry, dentry->d_inode, dir); | |
793 | else | |
2817b000 SW |
794 | d_drop(dentry); |
795 | return err; | |
796 | } | |
797 | ||
798 | static int ceph_link(struct dentry *old_dentry, struct inode *dir, | |
799 | struct dentry *dentry) | |
800 | { | |
3d14c5d2 YS |
801 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
802 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
803 | struct ceph_mds_request *req; |
804 | int err; | |
805 | ||
806 | if (ceph_snap(dir) != CEPH_NOSNAP) | |
807 | return -EROFS; | |
808 | ||
809 | dout("link in dir %p old_dentry %p dentry %p\n", dir, | |
810 | old_dentry, dentry); | |
811 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS); | |
812 | if (IS_ERR(req)) { | |
813 | d_drop(dentry); | |
814 | return PTR_ERR(req); | |
815 | } | |
816 | req->r_dentry = dget(dentry); | |
817 | req->r_num_caps = 2; | |
4b58c9b1 | 818 | req->r_old_dentry = dget(old_dentry); |
2817b000 SW |
819 | req->r_locked_dir = dir; |
820 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
821 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
ad88f23f YZ |
822 | /* release LINK_SHARED on source inode (mds will lock it) */ |
823 | req->r_old_inode_drop = CEPH_CAP_LINK_SHARED; | |
2817b000 | 824 | err = ceph_mdsc_do_request(mdsc, dir, req); |
70b666c3 | 825 | if (err) { |
2817b000 | 826 | d_drop(dentry); |
70b666c3 SW |
827 | } else if (!req->r_reply_info.head->is_dentry) { |
828 | ihold(old_dentry->d_inode); | |
829 | d_instantiate(dentry, old_dentry->d_inode); | |
830 | } | |
2817b000 SW |
831 | ceph_mdsc_put_request(req); |
832 | return err; | |
833 | } | |
834 | ||
835 | /* | |
836 | * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps. If it | |
837 | * looks like the link count will hit 0, drop any other caps (other | |
838 | * than PIN) we don't specifically want (due to the file still being | |
839 | * open). | |
840 | */ | |
841 | static int drop_caps_for_unlink(struct inode *inode) | |
842 | { | |
843 | struct ceph_inode_info *ci = ceph_inode(inode); | |
844 | int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
845 | ||
be655596 | 846 | spin_lock(&ci->i_ceph_lock); |
2817b000 SW |
847 | if (inode->i_nlink == 1) { |
848 | drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN); | |
849 | ci->i_ceph_flags |= CEPH_I_NODELAY; | |
850 | } | |
be655596 | 851 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
852 | return drop; |
853 | } | |
854 | ||
855 | /* | |
856 | * rmdir and unlink are differ only by the metadata op code | |
857 | */ | |
858 | static int ceph_unlink(struct inode *dir, struct dentry *dentry) | |
859 | { | |
3d14c5d2 YS |
860 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
861 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
862 | struct inode *inode = dentry->d_inode; |
863 | struct ceph_mds_request *req; | |
864 | int err = -EROFS; | |
865 | int op; | |
866 | ||
867 | if (ceph_snap(dir) == CEPH_SNAPDIR) { | |
868 | /* rmdir .snap/foo is RMSNAP */ | |
869 | dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len, | |
870 | dentry->d_name.name, dentry); | |
871 | op = CEPH_MDS_OP_RMSNAP; | |
872 | } else if (ceph_snap(dir) == CEPH_NOSNAP) { | |
873 | dout("unlink/rmdir dir %p dn %p inode %p\n", | |
874 | dir, dentry, inode); | |
dba19c60 | 875 | op = S_ISDIR(dentry->d_inode->i_mode) ? |
2817b000 SW |
876 | CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK; |
877 | } else | |
878 | goto out; | |
879 | req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS); | |
880 | if (IS_ERR(req)) { | |
881 | err = PTR_ERR(req); | |
882 | goto out; | |
883 | } | |
884 | req->r_dentry = dget(dentry); | |
885 | req->r_num_caps = 2; | |
886 | req->r_locked_dir = dir; | |
887 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
888 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
889 | req->r_inode_drop = drop_caps_for_unlink(inode); | |
890 | err = ceph_mdsc_do_request(mdsc, dir, req); | |
891 | if (!err && !req->r_reply_info.head->is_dentry) | |
892 | d_delete(dentry); | |
893 | ceph_mdsc_put_request(req); | |
894 | out: | |
895 | return err; | |
896 | } | |
897 | ||
898 | static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry, | |
899 | struct inode *new_dir, struct dentry *new_dentry) | |
900 | { | |
3d14c5d2 YS |
901 | struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb); |
902 | struct ceph_mds_client *mdsc = fsc->mdsc; | |
2817b000 SW |
903 | struct ceph_mds_request *req; |
904 | int err; | |
905 | ||
906 | if (ceph_snap(old_dir) != ceph_snap(new_dir)) | |
907 | return -EXDEV; | |
908 | if (ceph_snap(old_dir) != CEPH_NOSNAP || | |
909 | ceph_snap(new_dir) != CEPH_NOSNAP) | |
910 | return -EROFS; | |
911 | dout("rename dir %p dentry %p to dir %p dentry %p\n", | |
912 | old_dir, old_dentry, new_dir, new_dentry); | |
913 | req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS); | |
914 | if (IS_ERR(req)) | |
915 | return PTR_ERR(req); | |
180061a5 | 916 | ihold(old_dir); |
2817b000 SW |
917 | req->r_dentry = dget(new_dentry); |
918 | req->r_num_caps = 2; | |
919 | req->r_old_dentry = dget(old_dentry); | |
180061a5 | 920 | req->r_old_dentry_dir = old_dir; |
2817b000 SW |
921 | req->r_locked_dir = new_dir; |
922 | req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
923 | req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
924 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; | |
925 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; | |
926 | /* release LINK_RDCACHE on source inode (mds will lock it) */ | |
927 | req->r_old_inode_drop = CEPH_CAP_LINK_SHARED; | |
928 | if (new_dentry->d_inode) | |
929 | req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode); | |
930 | err = ceph_mdsc_do_request(mdsc, old_dir, req); | |
931 | if (!err && !req->r_reply_info.head->is_dentry) { | |
932 | /* | |
933 | * Normally d_move() is done by fill_trace (called by | |
934 | * do_request, above). If there is no trace, we need | |
935 | * to do it here. | |
936 | */ | |
ea1409f9 SW |
937 | |
938 | /* d_move screws up d_subdirs order */ | |
2f276c51 | 939 | ceph_dir_clear_complete(new_dir); |
ea1409f9 | 940 | |
2817b000 | 941 | d_move(old_dentry, new_dentry); |
ea1409f9 SW |
942 | |
943 | /* ensure target dentry is invalidated, despite | |
944 | rehashing bug in vfs_rename_dir */ | |
81a6cf2d | 945 | ceph_invalidate_dentry_lease(new_dentry); |
2817b000 SW |
946 | } |
947 | ceph_mdsc_put_request(req); | |
948 | return err; | |
949 | } | |
950 | ||
81a6cf2d SW |
951 | /* |
952 | * Ensure a dentry lease will no longer revalidate. | |
953 | */ | |
954 | void ceph_invalidate_dentry_lease(struct dentry *dentry) | |
955 | { | |
956 | spin_lock(&dentry->d_lock); | |
957 | dentry->d_time = jiffies; | |
958 | ceph_dentry(dentry)->lease_shared_gen = 0; | |
959 | spin_unlock(&dentry->d_lock); | |
960 | } | |
2817b000 SW |
961 | |
962 | /* | |
963 | * Check if dentry lease is valid. If not, delete the lease. Try to | |
964 | * renew if the least is more than half up. | |
965 | */ | |
966 | static int dentry_lease_is_valid(struct dentry *dentry) | |
967 | { | |
968 | struct ceph_dentry_info *di; | |
969 | struct ceph_mds_session *s; | |
970 | int valid = 0; | |
971 | u32 gen; | |
972 | unsigned long ttl; | |
973 | struct ceph_mds_session *session = NULL; | |
974 | struct inode *dir = NULL; | |
975 | u32 seq = 0; | |
976 | ||
977 | spin_lock(&dentry->d_lock); | |
978 | di = ceph_dentry(dentry); | |
3d8eb7a9 | 979 | if (di->lease_session) { |
2817b000 | 980 | s = di->lease_session; |
d8fb02ab | 981 | spin_lock(&s->s_gen_ttl_lock); |
2817b000 SW |
982 | gen = s->s_cap_gen; |
983 | ttl = s->s_cap_ttl; | |
d8fb02ab | 984 | spin_unlock(&s->s_gen_ttl_lock); |
2817b000 SW |
985 | |
986 | if (di->lease_gen == gen && | |
987 | time_before(jiffies, dentry->d_time) && | |
988 | time_before(jiffies, ttl)) { | |
989 | valid = 1; | |
990 | if (di->lease_renew_after && | |
991 | time_after(jiffies, di->lease_renew_after)) { | |
992 | /* we should renew */ | |
993 | dir = dentry->d_parent->d_inode; | |
994 | session = ceph_get_mds_session(s); | |
995 | seq = di->lease_seq; | |
996 | di->lease_renew_after = 0; | |
997 | di->lease_renew_from = jiffies; | |
998 | } | |
2817b000 SW |
999 | } |
1000 | } | |
1001 | spin_unlock(&dentry->d_lock); | |
1002 | ||
1003 | if (session) { | |
1004 | ceph_mdsc_lease_send_msg(session, dir, dentry, | |
1005 | CEPH_MDS_LEASE_RENEW, seq); | |
1006 | ceph_put_mds_session(session); | |
1007 | } | |
1008 | dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid); | |
1009 | return valid; | |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * Check if directory-wide content lease/cap is valid. | |
1014 | */ | |
1015 | static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry) | |
1016 | { | |
1017 | struct ceph_inode_info *ci = ceph_inode(dir); | |
1018 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
1019 | int valid = 0; | |
1020 | ||
be655596 | 1021 | spin_lock(&ci->i_ceph_lock); |
2817b000 SW |
1022 | if (ci->i_shared_gen == di->lease_shared_gen) |
1023 | valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1); | |
be655596 | 1024 | spin_unlock(&ci->i_ceph_lock); |
2817b000 SW |
1025 | dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n", |
1026 | dir, (unsigned)ci->i_shared_gen, dentry, | |
1027 | (unsigned)di->lease_shared_gen, valid); | |
1028 | return valid; | |
1029 | } | |
1030 | ||
1031 | /* | |
1032 | * Check if cached dentry can be trusted. | |
1033 | */ | |
0b728e19 | 1034 | static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags) |
2817b000 | 1035 | { |
bf1c6aca | 1036 | int valid = 0; |
34286d66 NP |
1037 | struct inode *dir; |
1038 | ||
0b728e19 | 1039 | if (flags & LOOKUP_RCU) |
34286d66 NP |
1040 | return -ECHILD; |
1041 | ||
1cd3935b SW |
1042 | dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry, |
1043 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode, | |
1044 | ceph_dentry(dentry)->offset); | |
2817b000 | 1045 | |
bf1c6aca SW |
1046 | dir = ceph_get_dentry_parent_inode(dentry); |
1047 | ||
2817b000 SW |
1048 | /* always trust cached snapped dentries, snapdir dentry */ |
1049 | if (ceph_snap(dir) != CEPH_NOSNAP) { | |
1050 | dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry, | |
1051 | dentry->d_name.len, dentry->d_name.name, dentry->d_inode); | |
bf1c6aca SW |
1052 | valid = 1; |
1053 | } else if (dentry->d_inode && | |
1054 | ceph_snap(dentry->d_inode) == CEPH_SNAPDIR) { | |
1055 | valid = 1; | |
1056 | } else if (dentry_lease_is_valid(dentry) || | |
1057 | dir_lease_is_valid(dir, dentry)) { | |
9215aeea YZ |
1058 | if (dentry->d_inode) |
1059 | valid = ceph_is_any_caps(dentry->d_inode); | |
1060 | else | |
1061 | valid = 1; | |
2817b000 | 1062 | } |
2817b000 | 1063 | |
bf1c6aca | 1064 | dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid"); |
9215aeea | 1065 | if (valid) { |
bf1c6aca | 1066 | ceph_dentry_lru_touch(dentry); |
9215aeea YZ |
1067 | } else { |
1068 | ceph_dir_clear_complete(dir); | |
bf1c6aca | 1069 | d_drop(dentry); |
9215aeea | 1070 | } |
bf1c6aca SW |
1071 | iput(dir); |
1072 | return valid; | |
2817b000 SW |
1073 | } |
1074 | ||
1075 | /* | |
147851d2 | 1076 | * Release our ceph_dentry_info. |
2817b000 | 1077 | */ |
147851d2 | 1078 | static void ceph_d_release(struct dentry *dentry) |
2817b000 SW |
1079 | { |
1080 | struct ceph_dentry_info *di = ceph_dentry(dentry); | |
2817b000 | 1081 | |
147851d2 | 1082 | dout("d_release %p\n", dentry); |
3d8eb7a9 SW |
1083 | ceph_dentry_lru_del(dentry); |
1084 | if (di->lease_session) | |
1085 | ceph_put_mds_session(di->lease_session); | |
1086 | kmem_cache_free(ceph_dentry_cachep, di); | |
1087 | dentry->d_fsdata = NULL; | |
2817b000 SW |
1088 | } |
1089 | ||
1090 | static int ceph_snapdir_d_revalidate(struct dentry *dentry, | |
0b728e19 | 1091 | unsigned int flags) |
2817b000 SW |
1092 | { |
1093 | /* | |
1094 | * Eventually, we'll want to revalidate snapped metadata | |
1095 | * too... probably... | |
1096 | */ | |
1097 | return 1; | |
1098 | } | |
1099 | ||
b58dc410 SW |
1100 | /* |
1101 | * When the VFS prunes a dentry from the cache, we need to clear the | |
1102 | * complete flag on the parent directory. | |
1103 | * | |
1104 | * Called under dentry->d_lock. | |
1105 | */ | |
1106 | static void ceph_d_prune(struct dentry *dentry) | |
1107 | { | |
774ac21d | 1108 | dout("ceph_d_prune %p\n", dentry); |
b58dc410 SW |
1109 | |
1110 | /* do we have a valid parent? */ | |
8842b3be | 1111 | if (IS_ROOT(dentry)) |
b58dc410 SW |
1112 | return; |
1113 | ||
2f276c51 | 1114 | /* if we are not hashed, we don't affect dir's completeness */ |
b58dc410 SW |
1115 | if (d_unhashed(dentry)) |
1116 | return; | |
2817b000 | 1117 | |
b58dc410 SW |
1118 | /* |
1119 | * we hold d_lock, so d_parent is stable, and d_fsdata is never | |
1120 | * cleared until d_release | |
1121 | */ | |
2f276c51 | 1122 | ceph_dir_clear_complete(dentry->d_parent->d_inode); |
b58dc410 | 1123 | } |
2817b000 SW |
1124 | |
1125 | /* | |
1126 | * read() on a dir. This weird interface hack only works if mounted | |
1127 | * with '-o dirstat'. | |
1128 | */ | |
1129 | static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size, | |
1130 | loff_t *ppos) | |
1131 | { | |
1132 | struct ceph_file_info *cf = file->private_data; | |
496ad9aa | 1133 | struct inode *inode = file_inode(file); |
2817b000 SW |
1134 | struct ceph_inode_info *ci = ceph_inode(inode); |
1135 | int left; | |
ae598083 | 1136 | const int bufsize = 1024; |
2817b000 | 1137 | |
3d14c5d2 | 1138 | if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT)) |
2817b000 SW |
1139 | return -EISDIR; |
1140 | ||
1141 | if (!cf->dir_info) { | |
ae598083 | 1142 | cf->dir_info = kmalloc(bufsize, GFP_NOFS); |
2817b000 SW |
1143 | if (!cf->dir_info) |
1144 | return -ENOMEM; | |
1145 | cf->dir_info_len = | |
ae598083 | 1146 | snprintf(cf->dir_info, bufsize, |
2817b000 SW |
1147 | "entries: %20lld\n" |
1148 | " files: %20lld\n" | |
1149 | " subdirs: %20lld\n" | |
1150 | "rentries: %20lld\n" | |
1151 | " rfiles: %20lld\n" | |
1152 | " rsubdirs: %20lld\n" | |
1153 | "rbytes: %20lld\n" | |
1154 | "rctime: %10ld.%09ld\n", | |
1155 | ci->i_files + ci->i_subdirs, | |
1156 | ci->i_files, | |
1157 | ci->i_subdirs, | |
1158 | ci->i_rfiles + ci->i_rsubdirs, | |
1159 | ci->i_rfiles, | |
1160 | ci->i_rsubdirs, | |
1161 | ci->i_rbytes, | |
1162 | (long)ci->i_rctime.tv_sec, | |
1163 | (long)ci->i_rctime.tv_nsec); | |
1164 | } | |
1165 | ||
1166 | if (*ppos >= cf->dir_info_len) | |
1167 | return 0; | |
1168 | size = min_t(unsigned, size, cf->dir_info_len-*ppos); | |
1169 | left = copy_to_user(buf, cf->dir_info + *ppos, size); | |
1170 | if (left == size) | |
1171 | return -EFAULT; | |
1172 | *ppos += (size - left); | |
1173 | return size - left; | |
1174 | } | |
1175 | ||
1176 | /* | |
1177 | * an fsync() on a dir will wait for any uncommitted directory | |
1178 | * operations to commit. | |
1179 | */ | |
02c24a82 JB |
1180 | static int ceph_dir_fsync(struct file *file, loff_t start, loff_t end, |
1181 | int datasync) | |
2817b000 | 1182 | { |
496ad9aa | 1183 | struct inode *inode = file_inode(file); |
2817b000 SW |
1184 | struct ceph_inode_info *ci = ceph_inode(inode); |
1185 | struct list_head *head = &ci->i_unsafe_dirops; | |
1186 | struct ceph_mds_request *req; | |
1187 | u64 last_tid; | |
1188 | int ret = 0; | |
1189 | ||
1190 | dout("dir_fsync %p\n", inode); | |
02c24a82 JB |
1191 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); |
1192 | if (ret) | |
1193 | return ret; | |
1194 | mutex_lock(&inode->i_mutex); | |
1195 | ||
2817b000 SW |
1196 | spin_lock(&ci->i_unsafe_lock); |
1197 | if (list_empty(head)) | |
1198 | goto out; | |
1199 | ||
1200 | req = list_entry(head->prev, | |
1201 | struct ceph_mds_request, r_unsafe_dir_item); | |
1202 | last_tid = req->r_tid; | |
1203 | ||
1204 | do { | |
1205 | ceph_mdsc_get_request(req); | |
1206 | spin_unlock(&ci->i_unsafe_lock); | |
2ff179e6 | 1207 | |
2817b000 SW |
1208 | dout("dir_fsync %p wait on tid %llu (until %llu)\n", |
1209 | inode, req->r_tid, last_tid); | |
1210 | if (req->r_timeout) { | |
1211 | ret = wait_for_completion_timeout( | |
1212 | &req->r_safe_completion, req->r_timeout); | |
1213 | if (ret > 0) | |
1214 | ret = 0; | |
1215 | else if (ret == 0) | |
1216 | ret = -EIO; /* timed out */ | |
1217 | } else { | |
1218 | wait_for_completion(&req->r_safe_completion); | |
1219 | } | |
2817b000 SW |
1220 | ceph_mdsc_put_request(req); |
1221 | ||
2ff179e6 | 1222 | spin_lock(&ci->i_unsafe_lock); |
2817b000 SW |
1223 | if (ret || list_empty(head)) |
1224 | break; | |
1225 | req = list_entry(head->next, | |
1226 | struct ceph_mds_request, r_unsafe_dir_item); | |
1227 | } while (req->r_tid < last_tid); | |
1228 | out: | |
1229 | spin_unlock(&ci->i_unsafe_lock); | |
02c24a82 JB |
1230 | mutex_unlock(&inode->i_mutex); |
1231 | ||
2817b000 SW |
1232 | return ret; |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * We maintain a private dentry LRU. | |
1237 | * | |
1238 | * FIXME: this needs to be changed to a per-mds lru to be useful. | |
1239 | */ | |
1240 | void ceph_dentry_lru_add(struct dentry *dn) | |
1241 | { | |
1242 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1243 | struct ceph_mds_client *mdsc; | |
2817b000 | 1244 | |
04a419f9 SW |
1245 | dout("dentry_lru_add %p %p '%.*s'\n", di, dn, |
1246 | dn->d_name.len, dn->d_name.name); | |
3d8eb7a9 SW |
1247 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
1248 | spin_lock(&mdsc->dentry_lru_lock); | |
1249 | list_add_tail(&di->lru, &mdsc->dentry_lru); | |
1250 | mdsc->num_dentry++; | |
1251 | spin_unlock(&mdsc->dentry_lru_lock); | |
2817b000 SW |
1252 | } |
1253 | ||
1254 | void ceph_dentry_lru_touch(struct dentry *dn) | |
1255 | { | |
1256 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1257 | struct ceph_mds_client *mdsc; | |
2817b000 | 1258 | |
1cd3935b SW |
1259 | dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn, |
1260 | dn->d_name.len, dn->d_name.name, di->offset); | |
3d8eb7a9 SW |
1261 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
1262 | spin_lock(&mdsc->dentry_lru_lock); | |
1263 | list_move_tail(&di->lru, &mdsc->dentry_lru); | |
1264 | spin_unlock(&mdsc->dentry_lru_lock); | |
2817b000 SW |
1265 | } |
1266 | ||
1267 | void ceph_dentry_lru_del(struct dentry *dn) | |
1268 | { | |
1269 | struct ceph_dentry_info *di = ceph_dentry(dn); | |
1270 | struct ceph_mds_client *mdsc; | |
1271 | ||
04a419f9 SW |
1272 | dout("dentry_lru_del %p %p '%.*s'\n", di, dn, |
1273 | dn->d_name.len, dn->d_name.name); | |
3d8eb7a9 SW |
1274 | mdsc = ceph_sb_to_client(dn->d_sb)->mdsc; |
1275 | spin_lock(&mdsc->dentry_lru_lock); | |
1276 | list_del_init(&di->lru); | |
1277 | mdsc->num_dentry--; | |
1278 | spin_unlock(&mdsc->dentry_lru_lock); | |
2817b000 SW |
1279 | } |
1280 | ||
6c0f3af7 SW |
1281 | /* |
1282 | * Return name hash for a given dentry. This is dependent on | |
1283 | * the parent directory's hash function. | |
1284 | */ | |
e5f86dc3 | 1285 | unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn) |
6c0f3af7 | 1286 | { |
6c0f3af7 SW |
1287 | struct ceph_inode_info *dci = ceph_inode(dir); |
1288 | ||
1289 | switch (dci->i_dir_layout.dl_dir_hash) { | |
1290 | case 0: /* for backward compat */ | |
1291 | case CEPH_STR_HASH_LINUX: | |
1292 | return dn->d_name.hash; | |
1293 | ||
1294 | default: | |
1295 | return ceph_str_hash(dci->i_dir_layout.dl_dir_hash, | |
1296 | dn->d_name.name, dn->d_name.len); | |
1297 | } | |
1298 | } | |
1299 | ||
2817b000 SW |
1300 | const struct file_operations ceph_dir_fops = { |
1301 | .read = ceph_read_dir, | |
77acfa29 | 1302 | .iterate = ceph_readdir, |
2817b000 SW |
1303 | .llseek = ceph_dir_llseek, |
1304 | .open = ceph_open, | |
1305 | .release = ceph_release, | |
1306 | .unlocked_ioctl = ceph_ioctl, | |
1307 | .fsync = ceph_dir_fsync, | |
1308 | }; | |
1309 | ||
1310 | const struct inode_operations ceph_dir_iops = { | |
1311 | .lookup = ceph_lookup, | |
1312 | .permission = ceph_permission, | |
1313 | .getattr = ceph_getattr, | |
1314 | .setattr = ceph_setattr, | |
1315 | .setxattr = ceph_setxattr, | |
1316 | .getxattr = ceph_getxattr, | |
1317 | .listxattr = ceph_listxattr, | |
1318 | .removexattr = ceph_removexattr, | |
7221fe4c | 1319 | .get_acl = ceph_get_acl, |
72466d0b | 1320 | .set_acl = ceph_set_acl, |
2817b000 SW |
1321 | .mknod = ceph_mknod, |
1322 | .symlink = ceph_symlink, | |
1323 | .mkdir = ceph_mkdir, | |
1324 | .link = ceph_link, | |
1325 | .unlink = ceph_unlink, | |
1326 | .rmdir = ceph_unlink, | |
1327 | .rename = ceph_rename, | |
1328 | .create = ceph_create, | |
2d83bde9 | 1329 | .atomic_open = ceph_atomic_open, |
2817b000 SW |
1330 | }; |
1331 | ||
52dfb8ac | 1332 | const struct dentry_operations ceph_dentry_ops = { |
2817b000 | 1333 | .d_revalidate = ceph_d_revalidate, |
147851d2 | 1334 | .d_release = ceph_d_release, |
b58dc410 | 1335 | .d_prune = ceph_d_prune, |
2817b000 SW |
1336 | }; |
1337 | ||
52dfb8ac | 1338 | const struct dentry_operations ceph_snapdir_dentry_ops = { |
2817b000 | 1339 | .d_revalidate = ceph_snapdir_d_revalidate, |
147851d2 | 1340 | .d_release = ceph_d_release, |
2817b000 SW |
1341 | }; |
1342 | ||
52dfb8ac | 1343 | const struct dentry_operations ceph_snap_dentry_ops = { |
147851d2 | 1344 | .d_release = ceph_d_release, |
b58dc410 | 1345 | .d_prune = ceph_d_prune, |
2817b000 | 1346 | }; |